alysis-code 0.13.0__py3-none-any.whl

alysis_code/profile_presets.py

@@ -0,0 +1,1834 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from urllib.parse import urlsplit
+
+from .llm.cache_capabilities import (
+    CACHE_STRATEGY_ANTHROPIC_CACHE_CONTROL,
+    CACHE_STRATEGY_GEMINI_EXPLICIT_CACHED_CONTENT,
+    CACHE_STRATEGY_IMPLICIT_PROVIDER,
+    CACHE_STRATEGY_MISTRAL_PROMPT_CACHE_KEY,
+    CACHE_STRATEGY_OPENAI_PROMPT_CACHE,
+    CACHE_STRATEGY_OPENROUTER_STICKY_SESSION,
+    CACHE_STRATEGY_QWEN_CACHE_CONTROL_BLOCKS,
+    CACHE_STRATEGY_XAI_CONVERSATION_HEADER,
+    CACHE_USAGE_SCHEMA_ANTHROPIC,
+    CACHE_USAGE_SCHEMA_GEMINI,
+    CACHE_USAGE_SCHEMA_OPENAI,
+    CACHE_USAGE_SCHEMA_PROVIDER,
+    OPENROUTER_SESSION_ID_FIELD,
+    XAI_CONVERSATION_ID_HEADER_FIELD,
+    CacheCapabilitySpec,
+)
+from .llm.protocols import (
+    ANTHROPIC_MESSAGES_PROTOCOL,
+    GEMINI_GENERATE_CONTENT_PROTOCOL,
+    GEMINI_INTERACTIONS_PROTOCOL,
+    OPENAI_COMPAT_PROTOCOL,
+    OPENAI_RESPONSES_PROTOCOL,
+)
+from .profiles import ProfileSpec
+from .web_search_adapters import (
+    ANTHROPIC_MESSAGES_ADAPTER,
+    AUTO_WEB_SEARCH_ADAPTER,
+    COHERE_WEB_SEARCH_ADAPTER,
+    DASHSCOPE_CHAT_ADAPTER,
+    GEMINI_GROUNDING_ADAPTER,
+    GROQ_COMPOUND_ADAPTER,
+    MINIMAX_CODING_PLAN_ADAPTER,
+    MISTRAL_CONVERSATIONS_ADAPTER,
+    MOONSHOT_KIMI_ADAPTER,
+    OPENAI_RESPONSES_ADAPTER,
+    OPENROUTER_WEB_ADAPTER,
+    PERPLEXITY_SONAR_ADAPTER,
+    VOLCENGINE_WEB_SEARCH_ADAPTER,
+    XAI_RESPONSES_ADAPTER,
+    ZHIPU_WEB_SEARCH_ADAPTER,
+)
+
+NATIVE_PROFILE_PROTOCOLS: frozenset[str] = frozenset(
+    {
+        OPENAI_RESPONSES_PROTOCOL,
+        ANTHROPIC_MESSAGES_PROTOCOL,
+        GEMINI_GENERATE_CONTENT_PROTOCOL,
+        GEMINI_INTERACTIONS_PROTOCOL,
+    }
+)
+FIRST_PARTY_NATIVE_PRESET_KEYS: tuple[str, ...] = (
+    "openai-responses",
+    "anthropic",
+    "gemini",
+)
+FIRST_CLASS_SETUP_PRESET_KEYS: tuple[str, ...] = (
+    # "alysis" (hosted MiMo) deliberately absent: while no campaign is
+    # running it stays off the primary picker entirely (advanced picker only).
+    *FIRST_PARTY_NATIVE_PRESET_KEYS,
+)
+FIRST_PARTY_COMPATIBILITY_PRESET_KEYS: tuple[str, ...] = (
+    "openai",
+    "anthropic-compat",
+    "gemini-compat",
+)
+LEGACY_NATIVE_ALIAS_PRESET_KEYS: tuple[str, ...] = ("anthropic-native", "gemini-native")
+LOCAL_PROFILE_PRESET_KEYS: tuple[str, ...] = ("ollama", "lm-studio", "vllm")
+_CUSTOM_PRESET_KEY = "custom"
+# Account-gated hosted presets (`alysis login`, no API key). Kept off the
+# primary provider picker while no hosted campaign is running.
+_ACCOUNT_GATED_PRESET_KEYS: tuple[str, ...] = ("alysis",)
+_CONVERSION_PRESET_BY_FAMILY: dict[str, dict[str, str]] = {
+    "openai": {"native": "openai-responses", "compatibility": "openai"},
+    "anthropic": {"native": "anthropic", "compatibility": "anthropic-compat"},
+    "gemini": {"native": "gemini", "compatibility": "gemini-compat"},
+}
+
+
+@dataclass(frozen=True)
+class ProfilePreset:
+    key: str
+    label: str
+    protocol: str
+    base_url: str
+    api_key_env: str | None
+    extra_headers: dict[str, str] = field(default_factory=dict)
+    suggested_models: tuple[str, ...] = ()
+    suggested_model_descriptions: dict[str, str] = field(default_factory=dict)
+    model_aliases: dict[str, str] = field(default_factory=dict)
+    validation_model: str = ""
+    web_search_adapter: str = AUTO_WEB_SEARCH_ADAPTER
+    web_search_model: str = ""
+    setup_warning: str = ""
+    notes: str = ""
+    cache_capability: CacheCapabilitySpec | None = None
+    # Keep new optional fields at the end so extensions using the legacy
+    # positional constructor continue to bind the sixth argument to headers.
+    provider_key: str = ""
+
+
+_OPENAI_PROMPT_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_OPENAI_PROMPT_CACHE,
+    enabled=True,
+    supports_prompt_cache_key=True,
+    supports_prompt_cache_retention=True,
+    reports_cache_read_tokens=True,
+    reports_cache_write_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_OPENAI,
+    min_cacheable_tokens=1024,
+    source="preset",
+)
+
+
+def _cache_minimum(tokens: int) -> CacheCapabilitySpec:
+    """A scoped override that only narrows the minimum cacheable prefix."""
+
+    return CacheCapabilitySpec(min_cacheable_tokens=tokens, source="preset")
+
+
+# Anthropic's minimum cacheable prefix is per-model and does not move
+# monotonically with the version number: the 5-generation flagships halved the
+# floor to 512 while Opus 4.6/4.5 and Haiku 4.5 still need 4096. A prefix under
+# the floor is silently not cached, so the write premium buys nothing — the
+# floor is what makes the request-shape report say so instead of guessing.
+_ANTHROPIC_CACHE_CONTROL_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_ANTHROPIC_CACHE_CONTROL,
+    enabled=True,
+    supports_cache_control=True,
+    reports_cache_read_tokens=True,
+    reports_cache_write_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_ANTHROPIC,
+    min_cacheable_tokens=1024,
+    model_family_overrides=(
+        ("claude-opus-5", _cache_minimum(512)),
+        ("claude-fable-5", _cache_minimum(512)),
+        ("claude-mythos-5", _cache_minimum(512)),
+        ("claude-mythos-preview", _cache_minimum(2048)),
+        ("claude-opus-4-7", _cache_minimum(2048)),
+        ("claude-opus-4-6", _cache_minimum(4096)),
+        ("claude-opus-4-5", _cache_minimum(4096)),
+        ("claude-haiku-4-5", _cache_minimum(4096)),
+    ),
+    source="preset",
+)
+_GEMINI_EXPLICIT_CACHED_CONTENT_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_GEMINI_EXPLICIT_CACHED_CONTENT,
+    enabled=True,
+    supports_explicit_cached_content=True,
+    reports_cache_read_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_GEMINI,
+    min_cacheable_tokens=4096,
+    source="preset",
+)
+_MISTRAL_PROMPT_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_MISTRAL_PROMPT_CACHE_KEY,
+    enabled=True,
+    supports_prompt_cache_key=True,
+    reports_cache_read_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_OPENAI,
+    min_cacheable_tokens=1024,
+    emits_request_fields=True,
+    notes=("Emits Mistral prompt_cache_key for stable server routing and prompt-cache hits.",),
+    source="preset",
+)
+_OPENROUTER_STICKY_SESSION_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_OPENROUTER_STICKY_SESSION,
+    enabled=True,
+    reports_cache_read_tokens=True,
+    reports_cache_write_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_PROVIDER,
+    emits_request_fields=True,
+    request_fields=(OPENROUTER_SESSION_ID_FIELD,),
+    notes=(
+        "Emits OpenRouter session_id for sticky routing; upstream cache semantics remain "
+        "route-dependent.",
+    ),
+    source="preset",
+)
+_XAI_CONVERSATION_HEADER_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_XAI_CONVERSATION_HEADER,
+    enabled=True,
+    reports_cache_read_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_PROVIDER,
+    emits_request_fields=True,
+    request_fields=(XAI_CONVERSATION_ID_HEADER_FIELD,),
+    notes=("Emits x-grok-conv-id for sticky cache routing on xAI Chat Completions.",),
+    source="preset",
+)
+_QWEN_DIAGNOSTIC_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_QWEN_CACHE_CONTROL_BLOCKS,
+    enabled=True,
+    reports_cache_read_tokens=True,
+    reports_cache_write_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_PROVIDER,
+    min_cacheable_tokens=1024,
+    emits_request_fields=False,
+    notes=(
+        "Diagnostic-only in auto mode; Qwen cache_control content markers mutate "
+        "message shape and require request-shape gating.",
+    ),
+    source="preset",
+)
+_MOONSHOT_AUTOMATIC_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_IMPLICIT_PROVIDER,
+    enabled=True,
+    supports_prompt_cache_key=True,
+    reports_cache_read_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_PROVIDER,
+    emits_request_fields=True,
+    notes=(
+        "Moonshot caches matching prompt prefixes automatically; prompt_cache_key keeps a "
+        "session on a stable cache-affinity route.",
+    ),
+    source="preset",
+)
+_ZAI_CODING_PLAN_CACHE_CAPABILITY = CacheCapabilitySpec(
+    strategy=CACHE_STRATEGY_IMPLICIT_PROVIDER,
+    enabled=True,
+    reports_cache_read_tokens=True,
+    usage_schema=CACHE_USAGE_SCHEMA_PROVIDER,
+    emits_request_fields=False,
+    notes=(
+        "Z.AI Coding Plan caches matching prompt prefixes automatically and reports "
+        "cached-input usage; Alysis Code emits no provider-specific cache fields.",
+    ),
+    source="preset",
+)
+
+
+def preset_protocol_kind(preset: ProfilePreset) -> str:
+    return "native" if preset.protocol in NATIVE_PROFILE_PROTOCOLS else "compatibility"
+
+
+def preset_protocol_summary(preset: ProfilePreset) -> str:
+    if preset.protocol in NATIVE_PROFILE_PROTOCOLS:
+        return (
+            f"Native first-party protocol: {preset.protocol} (recommended for first-party API keys)"
+        )
+    return "Compatibility protocol: OpenAI-compatible chat transport"
+
+
+def preset_selection_label(preset: ProfilePreset) -> str:
+    """Return a setup/config label that keeps protocol details out of the primary choice."""
+    if preset.key == "alysis":
+        return "Alysis Code Pro (hosted models) - requires login"
+    if preset.key == "openai-responses":
+        return "OpenAI - Native Responses"
+    if preset.key in {"anthropic", "anthropic-native"}:
+        return "Anthropic Claude - Native Messages"
+    if preset.key in {"gemini", "gemini-native"}:
+        return "Google Gemini - Native GenerateContent"
+    if preset.key == "openai":
+        return "OpenAI - Compatibility/gateway Chat Completions"
+    if preset.key == "anthropic-compat":
+        return "Anthropic Claude compatibility - legacy OpenAI-compatible"
+    if preset.key == "gemini-compat":
+        return "Google Gemini compatibility - legacy OpenAI-compatible"
+    if preset.key in LOCAL_PROFILE_PRESET_KEYS:
+        return f"{preset.label} - Local endpoint"
+    if preset.key == "custom":
+        return "Custom OpenAI-compatible endpoint"
+    return preset.label
+
+
+def _advanced_only_preset_keys() -> frozenset[str]:
+    """Preset keys deliberately kept off the primary provider picker.
+
+    Everything else in :data:`PROFILE_PRESETS` is a real hosted provider — the
+    native first-party APIs *and* the third-party API/gateway endpoints — and is
+    surfaced directly so users are not limited to the big-three brands. Only the
+    OpenAI-compatible duplicates of the native first-party providers, local
+    endpoints (Ollama/LM Studio/vLLM), the manual custom-URL entry, the
+    one-release legacy aliases, and the account-gated hosted MiMo preset
+    (no hosted campaign is running, so it is not a provider choice) stay
+    behind the advanced picker.
+    """
+    return frozenset(
+        {
+            _CUSTOM_PRESET_KEY,
+            *FIRST_PARTY_COMPATIBILITY_PRESET_KEYS,
+            *LOCAL_PROFILE_PRESET_KEYS,
+            *LEGACY_NATIVE_ALIAS_PRESET_KEYS,
+            *_ACCOUNT_GATED_PRESET_KEYS,
+        }
+    )
+
+
+def provider_selection_presets() -> list[ProfilePreset]:
+    """Presets shown directly on the primary provider picker.
+
+    Native first-party providers lead — the best defaults for new users —
+    followed by every other hosted provider in registration order.
+    Compatibility duplicates, local endpoints, the custom-URL entry,
+    one-release legacy aliases, and the account-gated hosted MiMo preset are
+    the only presets held back for the advanced picker, so the user sees the
+    full range of hosted providers up front instead of just
+    OpenAI/Anthropic/Gemini.
+    """
+    by_key = PRESET_BY_KEY
+    advanced = _advanced_only_preset_keys()
+    leading = [by_key[key] for key in FIRST_CLASS_SETUP_PRESET_KEYS if key in by_key]
+    leading_keys = {preset.key for preset in leading}
+    rest = [
+        preset
+        for preset in PROFILE_PRESETS
+        if preset.key not in advanced and preset.key not in leading_keys
+    ]
+    return [*leading, *rest]
+
+
+def advanced_provider_selection_presets() -> list[ProfilePreset]:
+    """Return the compatibility, local, custom, legacy alias, and account-gated presets.
+
+    These are exactly the presets held off the primary provider picker: the
+    OpenAI-compatible duplicates of the native first-party providers, local
+    endpoints (Ollama/LM Studio/vLLM), the manual custom-URL entry, the
+    one-release legacy aliases, and the account-gated hosted MiMo preset.
+    """
+    by_key = PRESET_BY_KEY
+    first_party_compat = [
+        by_key[key] for key in FIRST_PARTY_COMPATIBILITY_PRESET_KEYS if key in by_key
+    ]
+    local = [by_key[key] for key in LOCAL_PROFILE_PRESET_KEYS if key in by_key]
+    custom = [by_key[_CUSTOM_PRESET_KEY]] if _CUSTOM_PRESET_KEY in by_key else []
+    aliases = [by_key[key] for key in LEGACY_NATIVE_ALIAS_PRESET_KEYS if key in by_key]
+    account_gated = [by_key[key] for key in _ACCOUNT_GATED_PRESET_KEYS if key in by_key]
+    return [*first_party_compat, *local, *custom, *aliases, *account_gated]
+
+
+PROFILE_PRESETS: tuple[ProfilePreset, ...] = (
+    ProfilePreset(
+        key="openai",
+        provider_key="openai",
+        label="OpenAI",
+        protocol="openai_compat",
+        base_url="https://api.openai.com/v1",
+        api_key_env="OPENAI_API_KEY",
+        suggested_models=(
+            "gpt-5.6-terra",
+            "gpt-5.6-sol",
+            "gpt-5.6-luna",
+            "gpt-5.3-codex",
+            "gpt-5.4-mini",
+            "gpt-5.4-nano",
+        ),
+        suggested_model_descriptions={
+            "gpt-5.6-terra": "default - balanced 5.6 tier, 1.05M context",
+            "gpt-5.6-sol": "advanced - flagship 5.6 tier, 1.05M context",
+            "gpt-5.6-luna": "fast - low-cost 5.6 tier, full 1.05M context",
+            "gpt-5.3-codex": "coding - agentic codex model, 400K context",
+            "gpt-5.4-mini": "fallback - cheap tier for subagents, 400K",
+            "gpt-5.4-nano": "economy - cheapest live id, 400K context",
+        },
+        model_aliases={
+            "gpt-5.6": "gpt-5.6-sol",
+            "gpt-5-nano": "gpt-5.4-nano",
+            # 2026-07-23 shutdowns from OpenAI's deprecations page: codex and
+            # chat-latest ids remap to the still-callable gpt-5.5 tier.
+            "gpt-5-codex": "gpt-5.5",
+            "gpt-5.1-codex": "gpt-5.5",
+            "gpt-5.1-codex-max": "gpt-5.5",
+            "gpt-5.2-codex": "gpt-5.5",
+            "gpt-5.1-codex-mini": "gpt-5.4-mini",
+            "gpt-5-chat-latest": "gpt-5.5",
+            "gpt-5.1-chat-latest": "gpt-5.5",
+        },
+        validation_model="gpt-5.4-nano",
+        web_search_adapter=OPENAI_RESPONSES_ADAPTER,
+        cache_capability=_OPENAI_PROMPT_CACHE_CAPABILITY,
+        setup_warning=(
+            "gpt-5.6/5.4 reject tool calls with reasoning_effort other than "
+            "'none' on Chat Completions (and 5.6 defaults to 'medium') — for "
+            "agentic runs use the OpenAI Responses preset, or pin effort to "
+            "'none' here."
+        ),
+    ),
+    ProfilePreset(
+        key="openai-responses",
+        provider_key="openai",
+        label="OpenAI Responses",
+        protocol="openai_responses",
+        base_url="https://api.openai.com/v1",
+        api_key_env="OPENAI_API_KEY",
+        suggested_models=(
+            "gpt-5.6-terra",
+            "gpt-5.6-sol",
+            "gpt-5.6-luna",
+            "gpt-5.3-codex",
+            "gpt-5.4-mini",
+            "gpt-5.4-nano",
+        ),
+        suggested_model_descriptions={
+            "gpt-5.6-terra": "default - balanced 5.6 tier, 1.05M context",
+            "gpt-5.6-sol": "advanced - flagship 5.6 tier, 1.05M context",
+            "gpt-5.6-luna": "fast - low-cost 5.6 tier, full 1.05M context",
+            "gpt-5.3-codex": "coding - agentic codex model, 400K context",
+            "gpt-5.4-mini": "fallback - cheap tier for subagents, 400K",
+            "gpt-5.4-nano": "economy - cheapest live id, 400K context",
+        },
+        model_aliases={
+            "gpt-5.6": "gpt-5.6-sol",
+            "gpt-5-nano": "gpt-5.4-nano",
+            # 2026-07-23 shutdowns from OpenAI's deprecations page: codex and
+            # chat-latest ids remap to the still-callable gpt-5.5 tier.
+            "gpt-5-codex": "gpt-5.5",
+            "gpt-5.1-codex": "gpt-5.5",
+            "gpt-5.1-codex-max": "gpt-5.5",
+            "gpt-5.2-codex": "gpt-5.5",
+            "gpt-5.1-codex-mini": "gpt-5.4-mini",
+            "gpt-5-chat-latest": "gpt-5.5",
+            "gpt-5.1-chat-latest": "gpt-5.5",
+        },
+        validation_model="gpt-5.4-nano",
+        web_search_adapter=OPENAI_RESPONSES_ADAPTER,
+        cache_capability=_OPENAI_PROMPT_CACHE_CAPABILITY,
+        notes=(
+            "Native OpenAI Responses API chat with SSE streaming support. Use the OpenAI compat "
+            "preset to keep Chat Completions-compatible behavior."
+        ),
+    ),
+    ProfilePreset(
+        key="anthropic",
+        provider_key="anthropic",
+        label="Anthropic Claude",
+        protocol="anthropic_messages",
+        base_url="https://api.anthropic.com/v1",
+        api_key_env="ANTHROPIC_API_KEY",
+        suggested_models=(
+            "claude-sonnet-5",
+            "claude-opus-5",
+            "claude-fable-5",
+            "claude-haiku-4-5",
+            "claude-opus-4-8",
+            "claude-opus-4-7",
+        ),
+        suggested_model_descriptions={
+            "claude-sonnet-5": "default - 1M context, best speed/intelligence mix",
+            "claude-opus-5": "advanced - agentic coding + deep reasoning, 1M ctx",
+            "claude-fable-5": "reasoning - adaptive thinking always on, 1M ctx",
+            "claude-haiku-4-5": "fast - 200K context, lowest cost tier",
+            "claude-opus-4-8": "fallback - previous-generation opus, 1M context",
+            "claude-opus-4-7": "legacy - prior opus generation, 1M context",
+        },
+        model_aliases={
+            # claude-sonnet-4-6 moved to Anthropic's Legacy table; Sonnet 5 is
+            # newer and cheaper. Retired haiku ids remap to the 4.5 bare alias.
+            "claude-sonnet-4": "claude-sonnet-5",
+            "claude-sonnet-4-5": "claude-sonnet-5",
+            "claude-sonnet-4-6": "claude-sonnet-5",
+            "claude-4-sonnet": "claude-sonnet-5",
+            "claude-3-5-haiku-latest": "claude-haiku-4-5",
+            "claude-3-5-haiku-20241022": "claude-haiku-4-5",
+            "claude-opus-4.8": "claude-opus-4-8",
+            "claude-opus-4.7": "claude-opus-4-7",
+            "claude-opus-4-1": "claude-opus-4-8",
+            "claude-opus-4-6": "claude-opus-4-8",
+        },
+        validation_model="claude-haiku-4-5",
+        web_search_adapter=ANTHROPIC_MESSAGES_ADAPTER,
+        cache_capability=_ANTHROPIC_CACHE_CONTROL_CAPABILITY,
+        notes=(
+            "Native Anthropic Messages API chat with SSE streaming support. Compatibility mode "
+            "remains available as anthropic-compat for legacy OpenAI-compatible fallback."
+        ),
+    ),
+    ProfilePreset(
+        key="anthropic-compat",
+        provider_key="anthropic",
+        label="Anthropic Claude compatibility",
+        protocol="openai_compat",
+        base_url="https://api.anthropic.com/v1/",
+        api_key_env="ANTHROPIC_API_KEY",
+        suggested_models=(
+            "claude-sonnet-5",
+            "claude-opus-5",
+            "claude-fable-5",
+            "claude-haiku-4-5",
+            "claude-opus-4-8",
+            "claude-opus-4-7",
+        ),
+        suggested_model_descriptions={
+            "claude-sonnet-5": "default - 1M context, best speed/intelligence mix",
+            "claude-opus-5": "advanced - agentic coding + deep reasoning, 1M ctx",
+            "claude-fable-5": "reasoning - adaptive thinking always on, 1M ctx",
+            "claude-haiku-4-5": "fast - 200K context, lowest cost tier",
+            "claude-opus-4-8": "fallback - previous-generation opus, 1M context",
+            "claude-opus-4-7": "legacy - prior opus generation, 1M context",
+        },
+        model_aliases={
+            # claude-sonnet-4-6 moved to Anthropic's Legacy table; Sonnet 5 is
+            # newer and cheaper. Retired haiku ids remap to the 4.5 bare alias.
+            "claude-sonnet-4": "claude-sonnet-5",
+            "claude-sonnet-4-5": "claude-sonnet-5",
+            "claude-sonnet-4-6": "claude-sonnet-5",
+            "claude-4-sonnet": "claude-sonnet-5",
+            "claude-3-5-haiku-latest": "claude-haiku-4-5",
+            "claude-3-5-haiku-20241022": "claude-haiku-4-5",
+            "claude-opus-4.8": "claude-opus-4-8",
+            "claude-opus-4.7": "claude-opus-4-7",
+            "claude-opus-4-1": "claude-opus-4-8",
+            "claude-opus-4-6": "claude-opus-4-8",
+        },
+        validation_model="claude-haiku-4-5",
+        web_search_adapter=ANTHROPIC_MESSAGES_ADAPTER,
+        setup_warning=(
+            "Anthropic labels the OpenAI SDK compatibility layer as a test path; "
+            "use the anthropic preset for native Messages API behavior."
+        ),
+        notes=(
+            "Chat uses Anthropic OpenAI-compat at /v1; web_search uses the native "
+            "Anthropic Messages web_search adapter when the model/account supports it."
+        ),
+    ),
+    ProfilePreset(
+        key="anthropic-native",
+        provider_key="anthropic",
+        label="Anthropic Claude (native alias)",
+        protocol="anthropic_messages",
+        base_url="https://api.anthropic.com/v1",
+        api_key_env="ANTHROPIC_API_KEY",
+        suggested_models=(
+            "claude-sonnet-5",
+            "claude-opus-5",
+            "claude-fable-5",
+            "claude-haiku-4-5",
+            "claude-opus-4-8",
+            "claude-opus-4-7",
+        ),
+        suggested_model_descriptions={
+            "claude-sonnet-5": "default - 1M context, best speed/intelligence mix",
+            "claude-opus-5": "advanced - agentic coding + deep reasoning, 1M ctx",
+            "claude-fable-5": "reasoning - adaptive thinking always on, 1M ctx",
+            "claude-haiku-4-5": "fast - 200K context, lowest cost tier",
+            "claude-opus-4-8": "fallback - previous-generation opus, 1M context",
+            "claude-opus-4-7": "legacy - prior opus generation, 1M context",
+        },
+        model_aliases={
+            # claude-sonnet-4-6 moved to Anthropic's Legacy table; Sonnet 5 is
+            # newer and cheaper. Retired haiku ids remap to the 4.5 bare alias.
+            "claude-sonnet-4": "claude-sonnet-5",
+            "claude-sonnet-4-5": "claude-sonnet-5",
+            "claude-sonnet-4-6": "claude-sonnet-5",
+            "claude-4-sonnet": "claude-sonnet-5",
+            "claude-3-5-haiku-latest": "claude-haiku-4-5",
+            "claude-3-5-haiku-20241022": "claude-haiku-4-5",
+            "claude-opus-4.8": "claude-opus-4-8",
+            "claude-opus-4.7": "claude-opus-4-7",
+            "claude-opus-4-1": "claude-opus-4-8",
+            "claude-opus-4-6": "claude-opus-4-8",
+        },
+        validation_model="claude-haiku-4-5",
+        web_search_adapter=ANTHROPIC_MESSAGES_ADAPTER,
+        cache_capability=_ANTHROPIC_CACHE_CONTROL_CAPABILITY,
+        notes=(
+            "Legacy alias for the native anthropic preset. Prefer the anthropic preset for new "
+            "first-party Claude profiles."
+        ),
+    ),
+    ProfilePreset(
+        key="gemini",
+        provider_key="gemini",
+        label="Google Gemini",
+        protocol="gemini_generate_content",
+        base_url="https://generativelanguage.googleapis.com/v1beta",
+        api_key_env="GEMINI_API_KEY",
+        suggested_models=(
+            "gemini-3.7-flash",
+            "gemini-3.6-flash",
+            "gemini-3.5-flash-lite",
+            "gemini-3.1-pro-preview",
+        ),
+        suggested_model_descriptions={
+            "gemini-3.7-flash": "default - newest GA coding and agentic model, 1M",
+            "gemini-3.6-flash": "fallback - production GA flash model, 1M context",
+            "gemini-3.5-flash-lite": "economy - lowest-cost GA tier, 1M context",
+            "gemini-3.1-pro-preview": "advanced - pro reasoning preview, 1M context",
+        },
+        model_aliases={
+            # Only shut-down or invalid legacy ids are rewritten. Active stable
+            # ids and provider-managed *-latest aliases pass through unchanged.
+            "gemini-2.0-flash": "gemini-3.6-flash",
+            "gemini-2.0-flash-lite": "gemini-3.1-flash-lite",
+            "gemini-3.1-preview": "gemini-3.1-pro-preview",
+            "gemini-3-pro-preview": "gemini-3.1-pro-preview",
+            "gemini-3.1-flash-lite-preview": "gemini-3.1-flash-lite",
+        },
+        validation_model="gemini-3.5-flash-lite",
+        web_search_adapter=GEMINI_GROUNDING_ADAPTER,
+        cache_capability=_GEMINI_EXPLICIT_CACHED_CONTENT_CAPABILITY,
+        setup_warning=(
+            "Gemini native GenerateContent uses the Google Gemini API v1beta surface and "
+            "model availability can vary by account, region, and provider rollout."
+        ),
+        notes=(
+            "Native Gemini GenerateContent API chat with streamGenerateContent SSE support. "
+            "Compatibility mode remains available as gemini-compat for legacy OpenAI-compatible "
+            "fallback."
+        ),
+    ),
+    ProfilePreset(
+        key="gemini-compat",
+        provider_key="gemini",
+        label="Google Gemini compatibility",
+        protocol="openai_compat",
+        base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
+        api_key_env="GEMINI_API_KEY",
+        suggested_models=(
+            "gemini-3.7-flash",
+            "gemini-3.6-flash",
+            "gemini-3.5-flash-lite",
+            "gemini-3.1-pro-preview",
+        ),
+        suggested_model_descriptions={
+            "gemini-3.7-flash": "default - newest GA coding and agentic model, 1M",
+            "gemini-3.6-flash": "fallback - production GA flash model, 1M context",
+            "gemini-3.5-flash-lite": "economy - lowest-cost GA tier, 1M context",
+            "gemini-3.1-pro-preview": "advanced - pro reasoning preview, 1M context",
+        },
+        model_aliases={
+            # Only shut-down or invalid legacy ids are rewritten. Active stable
+            # ids and provider-managed *-latest aliases pass through unchanged.
+            "gemini-2.0-flash": "gemini-3.6-flash",
+            "gemini-2.0-flash-lite": "gemini-3.1-flash-lite",
+            "gemini-3.1-preview": "gemini-3.1-pro-preview",
+            "gemini-3-pro-preview": "gemini-3.1-pro-preview",
+            "gemini-3.1-flash-lite-preview": "gemini-3.1-flash-lite",
+        },
+        validation_model="gemini-3.5-flash-lite",
+        web_search_adapter=GEMINI_GROUNDING_ADAPTER,
+        setup_warning=(
+            "Gemini OpenAI compatibility is served from v1beta; use the gemini preset for "
+            "native GenerateContent behavior."
+        ),
+    ),
+    ProfilePreset(
+        key="gemini-native",
+        provider_key="gemini",
+        label="Google Gemini (native alias)",
+        protocol="gemini_generate_content",
+        base_url="https://generativelanguage.googleapis.com/v1beta",
+        api_key_env="GEMINI_API_KEY",
+        suggested_models=(
+            "gemini-3.7-flash",
+            "gemini-3.6-flash",
+            "gemini-3.5-flash-lite",
+            "gemini-3.1-pro-preview",
+        ),
+        suggested_model_descriptions={
+            "gemini-3.7-flash": "default - newest GA coding and agentic model, 1M",
+            "gemini-3.6-flash": "fallback - production GA flash model, 1M context",
+            "gemini-3.5-flash-lite": "economy - lowest-cost GA tier, 1M context",
+            "gemini-3.1-pro-preview": "advanced - pro reasoning preview, 1M context",
+        },
+        model_aliases={
+            # Only shut-down or invalid legacy ids are rewritten. Active stable
+            # ids and provider-managed *-latest aliases pass through unchanged.
+            "gemini-2.0-flash": "gemini-3.6-flash",
+            "gemini-2.0-flash-lite": "gemini-3.1-flash-lite",
+            "gemini-3.1-preview": "gemini-3.1-pro-preview",
+            "gemini-3-pro-preview": "gemini-3.1-pro-preview",
+            "gemini-3.1-flash-lite-preview": "gemini-3.1-flash-lite",
+        },
+        validation_model="gemini-3.5-flash-lite",
+        web_search_adapter=GEMINI_GROUNDING_ADAPTER,
+        cache_capability=_GEMINI_EXPLICIT_CACHED_CONTENT_CAPABILITY,
+        setup_warning=(
+            "Gemini native GenerateContent uses the Google Gemini API v1beta surface and "
+            "model availability can vary by account, region, and provider rollout."
+        ),
+        notes=(
+            "Legacy alias for the native gemini preset. Prefer the gemini preset for new "
+            "first-party Gemini profiles."
+        ),
+    ),
+    ProfilePreset(
+        key="deepseek",
+        provider_key="deepseek",
+        label="DeepSeek",
+        protocol="openai_compat",
+        base_url="https://api.deepseek.com",
+        api_key_env="DEEPSEEK_API_KEY",
+        suggested_models=(
+            "deepseek-v4-pro",
+            "deepseek-v4-flash",
+            "deepseek-v4-flash-vision-exp",
+        ),
+        suggested_model_descriptions={
+            "deepseek-v4-pro": "default - flagship coding model, 1M context",
+            "deepseek-v4-flash": "fast - cheap high-volume work, 1M context",
+            "deepseek-v4-flash-vision-exp": (
+                "vision preview - image understanding and tools, 1M context"
+            ),
+        },
+        model_aliases={
+            # deepseek-chat / deepseek-reasoner are discontinued 2026-07-24;
+            # saved configs pinning them keep working via these remaps.
+            "deepseek-chat": "deepseek-v4-flash",
+            "deepseek-reasoner": "deepseek-v4-flash",
+        },
+        validation_model="deepseek-v4-flash",
+        setup_warning=(
+            "Do not use retired legacy aliases deepseek-chat or deepseek-reasoner "
+            "for production defaults; use the V4 model IDs. The vision model is "
+            "experimental and may change without a stable-release deprecation window."
+        ),
+    ),
+    ProfilePreset(
+        key="nvidia",
+        provider_key="nvidia",
+        label="NVIDIA NIM (Hosted)",
+        protocol="openai_compat",
+        base_url="https://integrate.api.nvidia.com/v1",
+        api_key_env="NVIDIA_API_KEY",
+        suggested_models=(
+            "nvidia/nemotron-3-super-120b-a12b",
+            "nvidia/nemotron-3-ultra-550b-a55b",
+            "nvidia/nemotron-3-nano-30b-a3b",
+            "deepseek-ai/deepseek-v4-pro",
+            "deepseek-ai/deepseek-v4-flash",
+        ),
+        suggested_model_descriptions={
+            "nvidia/nemotron-3-super-120b-a12b": (
+                "default - balanced agentic reasoning, 1M context"
+            ),
+            "nvidia/nemotron-3-ultra-550b-a55b": (
+                "advanced - frontier agentic reasoning, 1M context"
+            ),
+            "nvidia/nemotron-3-nano-30b-a3b": (
+                "fast - efficient reasoning and tool use, 262K hosted context"
+            ),
+            "deepseek-ai/deepseek-v4-pro": (
+                "third-party model hosted by NVIDIA - advanced agentic reasoning"
+            ),
+            "deepseek-ai/deepseek-v4-flash": (
+                "third-party model hosted by NVIDIA - fast agentic reasoning"
+            ),
+        },
+        validation_model="nvidia/nemotron-3-nano-30b-a3b",
+        setup_warning=(
+            "NVIDIA hosted Free Endpoints are rate-limited development endpoints for "
+            "prototyping; availability is not a production SLA and may vary by account."
+        ),
+        notes=(
+            "Hosted NVIDIA NIM OpenAI-compatible API. The live catalog includes models "
+            "from NVIDIA and third parties; reasoning controls are model-specific."
+        ),
+    ),
+    ProfilePreset(
+        key="qwen-intl",
+        provider_key="qwen",
+        label="Alibaba Qwen / DashScope (Intl)",
+        protocol="openai_compat",
+        base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
+        api_key_env="DASHSCOPE_API_KEY",
+        suggested_models=(
+            "qwen3.7-plus",
+            "qwen3.8-max",
+            "qwen3.7-max",
+            "qwen3-coder-plus",
+            "qwen3-coder-next",
+            "qwen3.6-flash",
+            "qwen-flash",
+        ),
+        suggested_model_descriptions={
+            "qwen3.7-plus": "default - 1M context, balanced cost",
+            "qwen3.8-max": "advanced - newest multimodal flagship, 1M context",
+            "qwen3.7-max": "fallback - previous flagship, 1M context",
+            "qwen3-coder-plus": "coding - 1M context, long-repo work",
+            "qwen3-coder-next": "agentic - newest coder, 256K context",
+            "qwen3.6-flash": "fast - lower-latency, 1M context",
+            "qwen-flash": "economy - cheapest 1M-context option",
+        },
+        validation_model="qwen-flash",
+        web_search_adapter=DASHSCOPE_CHAT_ADAPTER,
+        web_search_model="qwen3.7-plus",
+        cache_capability=_QWEN_DIAGNOSTIC_CACHE_CAPABILITY,
+        setup_warning=(
+            "DashScope API keys are region-specific; use a key from the Singapore region."
+        ),
+    ),
+    ProfilePreset(
+        key="qwen-us",
+        provider_key="qwen",
+        label="Alibaba Qwen / DashScope (US)",
+        protocol="openai_compat",
+        base_url="https://dashscope-us.aliyuncs.com/compatible-mode/v1",
+        api_key_env="DASHSCOPE_API_KEY",
+        suggested_models=(
+            "qwen3.7-plus",
+            "qwen3.8-max",
+            "qwen3.7-max",
+            "qwen3.6-flash",
+            "qwen-flash",
+        ),
+        suggested_model_descriptions={
+            "qwen3.7-plus": "default - 1M context, balanced cost",
+            "qwen3.8-max": "advanced - newest multimodal flagship, 1M context",
+            "qwen3.7-max": "fallback - previous flagship, 1M context",
+            "qwen3.6-flash": "fast - lower-latency, 1M context",
+            "qwen-flash": "economy - cheapest 1M-context option",
+        },
+        validation_model="qwen-flash",
+        web_search_adapter=DASHSCOPE_CHAT_ADAPTER,
+        web_search_model="qwen3.7-plus",
+        cache_capability=_QWEN_DIAGNOSTIC_CACHE_CAPABILITY,
+        setup_warning=(
+            "DashScope API keys are region-specific; use a key from the US region. "
+            "Qwen coder models are not served from US (Virginia) — use qwen3.7-plus "
+            "for code work."
+        ),
+    ),
+    ProfilePreset(
+        key="qwen-cn",
+        provider_key="qwen",
+        label="Alibaba Qwen / DashScope (China)",
+        protocol="openai_compat",
+        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+        api_key_env="DASHSCOPE_API_KEY",
+        suggested_models=(
+            "qwen3.7-plus",
+            "qwen3.8-max",
+            "qwen3.7-max",
+            "qwen3-coder-plus",
+            "qwen3-coder-next",
+            "qwen3.6-flash",
+            "qwen-flash",
+        ),
+        suggested_model_descriptions={
+            "qwen3.7-plus": "default - 1M context, balanced cost",
+            "qwen3.8-max": "advanced - newest multimodal flagship, 1M context",
+            "qwen3.7-max": "fallback - previous flagship, 1M context",
+            "qwen3-coder-plus": "coding - 1M context, long-repo work",
+            "qwen3-coder-next": "agentic - newest coder, 256K context",
+            "qwen3.6-flash": "fast - lower-latency, 1M context",
+            "qwen-flash": "economy - cheapest 1M-context option",
+        },
+        validation_model="qwen-flash",
+        web_search_adapter=DASHSCOPE_CHAT_ADAPTER,
+        web_search_model="qwen3.7-plus",
+        cache_capability=_QWEN_DIAGNOSTIC_CACHE_CAPABILITY,
+        setup_warning="DashScope API keys are region-specific; use a key from the China region.",
+    ),
+    ProfilePreset(
+        key="zhipu",
+        provider_key="zhipu",
+        label="Zhipu / GLM",
+        protocol="openai_compat",
+        base_url="https://open.bigmodel.cn/api/paas/v4/",
+        api_key_env="ZHIPUAI_API_KEY",
+        suggested_models=(
+            "glm-5.2",
+            "glm-5.1",
+            "glm-5-turbo",
+            "glm-4.7",
+            "glm-4.7-flashx",
+            "glm-4.7-flash",
+        ),
+        suggested_model_descriptions={
+            "glm-5.2": "default - 1M context, agentic coding",
+            "glm-5.1": "advanced - previous flagship, 200K context",
+            "glm-5-turbo": "coding - 200K context, cheaper than glm-5.1",
+            "glm-4.7": "fallback - cheap 200K context",
+            "glm-4.7-flashx": "fast - 200K context, no free-tier rate caps",
+            "glm-4.7-flash": "economy - free tier, 200K context, rate limited",
+        },
+        # No aliases on purpose: glm-5, glm-4.6 etc. remain individually priced
+        # and callable — remapping would silently change what users are billed.
+        validation_model="glm-4.7-flash",
+        web_search_adapter=ZHIPU_WEB_SEARCH_ADAPTER,
+        web_search_model="glm-5.1",
+    ),
+    ProfilePreset(
+        key="zai-coding-plan",
+        provider_key="zai_coding_plan",
+        label="Z.AI GLM Coding Plan",
+        protocol="openai_compat",
+        base_url="https://api.z.ai/api/coding/paas/v4",
+        api_key_env="ZAI_API_KEY",
+        suggested_models=(
+            "glm-5.3",
+            "glm-5-turbo",
+            "glm-4.7",
+        ),
+        suggested_model_descriptions={
+            "glm-5.3": "default - latest agentic coding model, 1M context",
+            "glm-5-turbo": "fast - lower-credit agent model, 200K context",
+            "glm-4.7": "fallback - lowest-credit plan model, 200K context",
+        },
+        # GLM-4.7 consumes fewer plan credits than GLM-5.3 and is available on
+        # every Coding Plan tier, so use it for the initial credential probe.
+        validation_model="glm-4.7",
+        cache_capability=_ZAI_CODING_PLAN_CACHE_CAPABILITY,
+        setup_warning=(
+            "Requires a Z.AI GLM Coding Plan key; general pay-as-you-go and "
+            "open.bigmodel.cn keys use different endpoints. Z.AI limits plan benefits "
+            "to supported coding tools, so verify Alysis Code eligibility for your account."
+        ),
+        notes=(
+            "Subscription Coding Plan endpoint, not the general Z.AI or China Zhipu API. "
+            "All plan tiers currently offer GLM-5.3, GLM-5-Turbo, and GLM-4.7; "
+            "GLM-5.2/5.1 requests are routed by the server to GLM-5.3."
+        ),
+    ),
+    ProfilePreset(
+        key="moonshot",
+        provider_key="moonshot",
+        label="Kimi",
+        protocol="openai_compat",
+        base_url="https://api.moonshot.ai/v1",
+        api_key_env="MOONSHOT_API_KEY",
+        suggested_models=(
+            "kimi-k2.7-code",
+            "kimi-k3",
+            "kimi-k2.7-code-highspeed",
+            "kimi-k2.6",
+        ),
+        suggested_model_descriptions={
+            # k2.7-code is the deliberate default: k3 is always-thinking at
+            # pinned max effort and ~3x the input price — escalate to it when a
+            # task needs the 1M window, don't route routine turns through it.
+            "kimi-k2.7-code": "default - 256K context, long-horizon agentic coding",
+            "kimi-k3": "advanced - 1M context, always-thinking at max effort",
+            "kimi-k2.7-code-highspeed": "fast - ~180 tok/s coding variant, 256K context",
+            "kimi-k2.6": "fallback - 256K context, thinking toggleable",
+        },
+        model_aliases={
+            # Only kimi-k2.6 accepts a thinking-off flag; K2.7/K3 error on it.
+            # kimi-k2.5 and the moonshot-v1-* family end 2026-08-31.
+            "kimi-k2": "kimi-k2.6",
+            "kimi-k2.5": "kimi-k2.6",
+            "kimi-k2-thinking": "kimi-k2.6",
+            "kimi-k2-thinking-turbo": "kimi-k2.6",
+            "kimi-k2-0905-preview": "kimi-k2.6",
+            "kimi-k2-0711-preview": "kimi-k2.6",
+            "kimi-k2-turbo-preview": "kimi-k2.7-code-highspeed",
+            "kimi-latest": "kimi-k2.6",
+            "kimi-thinking-preview": "kimi-k2.6",
+            "moonshot-v1-8k": "kimi-k2.6",
+            "moonshot-v1-32k": "kimi-k2.6",
+            "moonshot-v1-128k": "kimi-k2.6",
+            "moonshot-v1-auto": "kimi-k2.6",
+        },
+        validation_model="kimi-k2.6",
+        web_search_adapter=MOONSHOT_KIMI_ADAPTER,
+        # kimi-k3 cannot disable thinking, which Kimi's $web_search tool requires,
+        # so provider-hosted search stays pinned to kimi-k2.6.
+        web_search_model="kimi-k2.6",
+        cache_capability=_MOONSHOT_AUTOMATIC_CACHE_CAPABILITY,
+        setup_warning=(
+            "Moonshot API keys are region-scoped; use a key from the international "
+            "platform (platform.kimi.ai) with this endpoint."
+        ),
+    ),
+    ProfilePreset(
+        key="kimi-code",
+        provider_key="moonshot",
+        label="Kimi Code",
+        protocol="openai_compat",
+        base_url="https://api.kimi.com/coding/v1",
+        api_key_env="KIMI_API_KEY",
+        suggested_models=("k3", "kimi-for-coding", "kimi-for-coding-highspeed"),
+        suggested_model_descriptions={
+            # Tier gating: kimi-for-coding = all members; k3 = Moderato+ (256K)
+            # and 1M only on Allegretto+; -highspeed = Allegretto+ only.
+            # Disabling thinking on this endpoint silently routes to K2.6.
+            "k3": "default - 256K context, 1M on Allegretto+",
+            "kimi-for-coding": "coding - 256K context, all membership tiers",
+            "kimi-for-coding-highspeed": "fast - 256K context, Allegretto tier or above",
+        },
+        model_aliases={
+            # Cross-endpoint remaps: these are live, DIFFERENT ids on
+            # platform.moonshot.ai — legal only inside this preset's alias table.
+            "kimi-k3": "k3",
+            "kimi-k2.7-code": "kimi-for-coding",
+            "kimi-k2.7-code-highspeed": "kimi-for-coding-highspeed",
+        },
+        # Validation is a billed call against metered membership quota — no
+        # /models endpoint exists on this surface.
+        validation_model="kimi-for-coding",
+        setup_warning=(
+            "Requires a Kimi membership key from the kimi.com console; "
+            "platform.kimi.ai pay-as-you-go keys are not valid here. "
+            "Turning reasoning off routes requests to K2.6 (a different model)."
+        ),
+    ),
+    ProfilePreset(
+        key="moonshot-cn",
+        provider_key="moonshot",
+        label="Kimi (China)",
+        protocol="openai_compat",
+        base_url="https://api.moonshot.cn/v1",
+        api_key_env="MOONSHOT_API_KEY",
+        suggested_models=(
+            "kimi-k2.7-code",
+            "kimi-k3",
+            "kimi-k2.7-code-highspeed",
+            "kimi-k2.6",
+        ),
+        suggested_model_descriptions={
+            # k2.7-code is the deliberate default: k3 is always-thinking at
+            # pinned max effort and ~3x the input price — escalate to it when a
+            # task needs the 1M window, don't route routine turns through it.
+            "kimi-k2.7-code": "default - 256K context, long-horizon agentic coding",
+            "kimi-k3": "advanced - 1M context, always-thinking at max effort",
+            "kimi-k2.7-code-highspeed": "fast - ~180 tok/s coding variant, 256K context",
+            "kimi-k2.6": "fallback - 256K context, thinking toggleable",
+        },
+        model_aliases={
+            # Only kimi-k2.6 accepts a thinking-off flag; K2.7/K3 error on it.
+            # kimi-k2.5 and the moonshot-v1-* family end 2026-08-31.
+            "kimi-k2": "kimi-k2.6",
+            "kimi-k2.5": "kimi-k2.6",
+            "kimi-k2-thinking": "kimi-k2.6",
+            "kimi-k2-thinking-turbo": "kimi-k2.6",
+            "kimi-k2-0905-preview": "kimi-k2.6",
+            "kimi-k2-0711-preview": "kimi-k2.6",
+            "kimi-k2-turbo-preview": "kimi-k2.7-code-highspeed",
+            "kimi-latest": "kimi-k2.6",
+            "kimi-thinking-preview": "kimi-k2.6",
+            "moonshot-v1-8k": "kimi-k2.6",
+            "moonshot-v1-32k": "kimi-k2.6",
+            "moonshot-v1-128k": "kimi-k2.6",
+            "moonshot-v1-auto": "kimi-k2.6",
+        },
+        validation_model="kimi-k2.6",
+        web_search_adapter=MOONSHOT_KIMI_ADAPTER,
+        # kimi-k3 cannot disable thinking, which Kimi's $web_search tool requires,
+        # so provider-hosted search stays pinned to kimi-k2.6.
+        web_search_model="kimi-k2.6",
+        cache_capability=_MOONSHOT_AUTOMATIC_CACHE_CAPABILITY,
+        setup_warning=(
+            "Moonshot API keys are region-scoped; use a key from the mainland-China "
+            "platform (platform.moonshot.cn) with this endpoint."
+        ),
+    ),
+    ProfilePreset(
+        key="minimax",
+        provider_key="minimax",
+        label="MiniMax",
+        protocol="openai_compat",
+        base_url="https://api.minimax.io/v1",
+        api_key_env="MINIMAX_API_KEY",
+        suggested_models=(
+            "MiniMax-M3",
+            "MiniMax-M2.7",
+            "MiniMax-M2.7-highspeed",
+            "MiniMax-M2.5",
+        ),
+        suggested_model_descriptions={
+            # No thinking/reasoning toggle is documented for any MiniMax model —
+            # send no reasoning-control parameter on this preset. M3 input above
+            # 512K bills at a higher long-context rate.
+            "MiniMax-M3": "default - 1M context, multimodal agentic coding",
+            "MiniMax-M2.7": "coding - 200K context, prior flagship",
+            "MiniMax-M2.7-highspeed": "fast - same weights as M2.7, latency-tuned",
+            "MiniMax-M2.5": "fallback - stable prior generation",
+        },
+        model_aliases={
+            "MiniMax-M2": "MiniMax-M2.7",
+        },
+        validation_model="MiniMax-M2.5",
+        web_search_adapter=MINIMAX_CODING_PLAN_ADAPTER,
+        setup_warning=(
+            "MiniMax hosted web search requires a Token Plan key; pay-as-you-go model keys "
+            "cannot call the Token Plan search endpoint."
+        ),
+        notes=(
+            "Chat uses the OpenAI-compatible MiniMax API. Web search uses MiniMax's Token Plan "
+            "search endpoint when the configured key has Token Plan access."
+        ),
+    ),
+    ProfilePreset(
+        key="xiaomi-mimo",
+        provider_key="xiaomi",
+        label="Xiaomi MiMo",
+        protocol="openai_compat",
+        base_url="https://api.xiaomimimo.com/v1",
+        api_key_env="XIAOMI_API_KEY",
+        suggested_models=("mimo-v2.5-pro", "mimo-v2-flash", "mimo-v2.5"),
+        suggested_model_descriptions={
+            "mimo-v2.5-pro": "default - flagship reasoning, coding & agents (1M context)",
+            "mimo-v2-flash": "faster & lighter (256K context)",
+            "mimo-v2.5": "omni - text + image understanding (1M context)",
+        },
+        validation_model="mimo-v2.5-pro",
+        # Migrate the legacy bare "mimo" placeholder up to the flagship model.
+        model_aliases={"mimo": "mimo-v2.5-pro"},
+    ),
+    ProfilePreset(
+        key="bytedance",
+        provider_key="bytedance",
+        label="ByteDance Doubao",
+        protocol="openai_compat",
+        base_url="https://ark.cn-beijing.volces.com/api/v3",
+        api_key_env="ARK_API_KEY",
+        suggested_models=(
+            "doubao-seed-2-0-pro-260215",
+            "doubao-seed-2-0-code-preview-260215",
+            "doubao-seed-2-0-lite-260215",
+            "doubao-seed-2-0-mini-260215",
+        ),
+        suggested_model_descriptions={
+            "doubao-seed-2-0-pro-260215": "default - flagship seed 2.0, agentic tasks",
+            "doubao-seed-2-0-code-preview-260215": "coding - 256K context, preview snapshot",
+            "doubao-seed-2-0-lite-260215": "fast - balanced quality and latency",
+            "doubao-seed-2-0-mini-260215": "economy - cheapest seed 2.0, high concurrency",
+        },
+        validation_model="doubao-seed-2-0-mini-260215",
+        web_search_adapter=VOLCENGINE_WEB_SEARCH_ADAPTER,
+        setup_warning=(
+            "Model ids rest on registry evidence only (Ark docs are not "
+            "machine-readable) — verify with a live Ark key; Ark may require "
+            "endpoint ids (ep-...) instead of bare model names."
+        ),
+    ),
+    ProfilePreset(
+        key="groq",
+        provider_key="groq",
+        label="Groq",
+        protocol="openai_compat",
+        base_url="https://api.groq.com/openai/v1",
+        api_key_env="GROQ_API_KEY",
+        suggested_models=(
+            "openai/gpt-oss-120b",
+            "qwen/qwen3.6-27b",
+            "openai/gpt-oss-20b",
+            "groq/compound",
+        ),
+        suggested_model_descriptions={
+            # groq/compound runs server-side built-in tools and does NOT accept
+            # client tool_call — never route normal agent tool loops to it.
+            "openai/gpt-oss-120b": "default - 131K context, adjustable reasoning",
+            "qwen/qwen3.6-27b": "coding - thinking modes and vision, preview tier",
+            "openai/gpt-oss-20b": "fast - cheapest non-deprecated production id",
+            "groq/compound": "agentic - server-side web search and code exec",
+        },
+        model_aliases={
+            # Both llama ids shut down 2026-08-16 (Groq deprecations table);
+            # the other retired ids remap per the same table.
+            "llama-3.3-70b-versatile": "openai/gpt-oss-120b",
+            "llama-3.1-8b-instant": "openai/gpt-oss-20b",
+            "qwen/qwen3-32b": "openai/gpt-oss-120b",
+            "meta-llama/llama-4-scout-17b-16e-instruct": "openai/gpt-oss-120b",
+            "meta-llama/llama-4-maverick-17b-128e-instruct": "openai/gpt-oss-120b",
+            "moonshotai/kimi-k2-instruct": "openai/gpt-oss-120b",
+            "moonshotai/kimi-k2-instruct-0905": "openai/gpt-oss-120b",
+        },
+        validation_model="openai/gpt-oss-20b",
+        web_search_adapter=GROQ_COMPOUND_ADAPTER,
+        web_search_model="groq/compound-mini",
+        setup_warning=(
+            "Groq is mostly OpenAI-compatible; avoid preview-only models as production "
+            "defaults (qwen/qwen3.6-27b is preview and may be pulled without notice)."
+        ),
+    ),
+    ProfilePreset(
+        key="cerebras",
+        provider_key="cerebras",
+        label="Cerebras",
+        protocol="openai_compat",
+        base_url="https://api.cerebras.ai/v1",
+        api_key_env="CEREBRAS_API_KEY",
+        suggested_models=(
+            "gpt-oss-120b",
+            "zai-glm-4.7",
+            "gemma-4-31b",
+        ),
+        suggested_model_descriptions={
+            # Context values are the free-tier floor (65K); paid keys get 131K.
+            # gpt-oss-120b cannot disable reasoning (effort low|medium|high).
+            "gpt-oss-120b": "default - only GA public model, ~3000 tok/s",
+            "zai-glm-4.7": "coding - strongest here, deprecates 2026-08-17",
+            "gemma-4-31b": "fallback - only image-input model, preview tier",
+        },
+        model_aliases={
+            # The llama family left Cerebras public endpoints 2026-02-16 (and
+            # "llama3.3-70b" was never a valid spelling of the id).
+            "llama3.3-70b": "gpt-oss-120b",
+            "llama-3.3-70b": "gpt-oss-120b",
+            "llama3.1-70b": "gpt-oss-120b",
+            "llama3.1-8b": "gpt-oss-120b",
+            "qwen-3-32b": "gpt-oss-120b",
+            "qwen-3-coder-480b": "zai-glm-4.7",
+            "zai-glm-4.6": "zai-glm-4.7",
+            "deepseek-r1-distill-llama-70b": "gpt-oss-120b",
+        },
+        validation_model="gpt-oss-120b",
+    ),
+    ProfilePreset(
+        key="mistral",
+        provider_key="mistral",
+        label="Mistral AI",
+        protocol="openai_compat",
+        base_url="https://api.mistral.ai/v1",
+        api_key_env="MISTRAL_API_KEY",
+        suggested_models=(
+            "mistral-medium-3-5",
+            "mistral-large-2512",
+            "mistral-small-2603",
+            "codestral-2508",
+            "ministral-8b-2512",
+        ),
+        suggested_model_descriptions={
+            # codestral is FIM/completion-oriented with ~4K max output — routers
+            # should prefer the default for multi-file agentic patch turns.
+            "mistral-medium-3-5": "default - agentic and coding flagship, 256K",
+            "mistral-large-2512": "advanced - mistral large 3, 675B MoE, 256K",
+            "mistral-small-2603": "fast - mistral small 4, low latency",
+            "codestral-2508": "coding - FIM and completion, 4K max output",
+            "ministral-8b-2512": "economy - small tool-capable model",
+        },
+        model_aliases={
+            # Mistral documents mistral-medium-3-5 as the primary API id. Keep
+            # the former Alysis Code default as a compatibility alias, while the
+            # provider-managed -latest alias passes through unchanged.
+            "mistral-medium-2604": "mistral-medium-3-5",
+            "mistral-medium-3": "mistral-medium-3-5",
+            "mistral-medium-2508": "mistral-medium-3-5",
+            "mistral-medium-2505": "mistral-medium-3-5",
+            "mistral-small-latest": "mistral-small-2603",
+            "mistral-small-2506": "mistral-small-2603",
+            "mistral-large-latest": "mistral-large-2512",
+            "mistral-large-2411": "mistral-medium-3-5",
+            "mistral-large-2407": "mistral-large-2512",
+            "codestral-latest": "codestral-2508",
+            "devstral-2512": "mistral-medium-3-5",
+            "devstral-latest": "mistral-medium-3-5",
+            "devstral-medium-latest": "mistral-medium-3-5",
+            "devstral-medium-2507": "mistral-medium-3-5",
+            "devstral-small-2507": "mistral-small-2603",
+            "labs-devstral-small-2512": "mistral-medium-3-5",
+            "magistral-medium-latest": "mistral-medium-3-5",
+            "magistral-small-latest": "mistral-small-2603",
+            "ministral-8b-latest": "ministral-8b-2512",
+            "open-mistral-nemo-2407": "ministral-8b-2512",
+        },
+        validation_model="ministral-3b-2512",
+        web_search_adapter=MISTRAL_CONVERSATIONS_ADAPTER,
+        web_search_model="mistral-medium-latest",
+        cache_capability=_MISTRAL_PROMPT_CACHE_CAPABILITY,
+    ),
+    ProfilePreset(
+        key="xai",
+        provider_key="xai",
+        label="xAI Grok",
+        protocol="openai_compat",
+        base_url="https://api.x.ai/v1",
+        api_key_env="XAI_API_KEY",
+        suggested_models=(
+            "grok-4.6",
+            "grok-4.5",
+            "grok-build-0.1",
+            "grok-4.3",
+            "grok-4.20-0309-reasoning",
+            "grok-4.20-0309-non-reasoning",
+        ),
+        suggested_model_descriptions={
+            # grok-build-0.1 is served from us-east-1/us-west-2 only. Max output
+            # is unpublished for the 4.20 family — clamp conservatively.
+            "grok-4.6": "default - newest flagship for coding and agents, 500K",
+            "grok-4.5": "fallback - previous flagship for coding and agents",
+            "grok-build-0.1": "coding - agentic engineering model, 256K",
+            "grok-4.3": "advanced - 1M context window",
+            "grok-4.20-0309-reasoning": "reasoning - dedicated snapshot, 1M context",
+            "grok-4.20-0309-non-reasoning": "fast - no-reasoning snapshot, 1M context",
+        },
+        model_aliases={
+            # Retired 2026-05-15, full shutdown 2026-08-15. The *-non-reasoning
+            # slugs deliberately map to the non-reasoning snapshot (xAI's own
+            # redirect lands them on grok-4.3 with effort=none, which the alias
+            # table cannot express).
+            "grok-code-fast-1": "grok-build-0.1",
+            "grok-4": "grok-4.3",
+            "grok-4-0709": "grok-4.3",
+            "grok-4-fast": "grok-4.3",
+            "grok-4.3-latest": "grok-4.3",
+            "grok-4-fast-reasoning": "grok-4.3",
+            "grok-4-1-fast-reasoning": "grok-4.3",
+            "grok-4-fast-non-reasoning": "grok-4.20-0309-non-reasoning",
+            "grok-4-1-fast-non-reasoning": "grok-4.20-0309-non-reasoning",
+            "grok-3": "grok-4.3",
+        },
+        validation_model="grok-4.20-0309-non-reasoning",
+        web_search_adapter=XAI_RESPONSES_ADAPTER,
+        cache_capability=_XAI_CONVERSATION_HEADER_CACHE_CAPABILITY,
+        setup_warning=(
+            "Retired slugs (grok-4, grok-4-fast, grok-3, grok-code-fast-1) shut down "
+            "fully 2026-08-15 and are billed at grok-4.3 rates until then; migrate "
+            "pinned configs explicitly. Ids use dots, not dashes (grok-4.6)."
+        ),
+    ),
+    ProfilePreset(
+        key="cohere",
+        provider_key="cohere",
+        label="Cohere (compat)",
+        protocol="openai_compat",
+        base_url="https://api.cohere.ai/compatibility/v1",
+        api_key_env="COHERE_API_KEY",
+        suggested_models=(
+            "command-a-plus-05-2026",
+            "command-a-reasoning-08-2025",
+            "command-a-03-2025",
+            "command-r7b-12-2024",
+        ),
+        suggested_model_descriptions={
+            # Reasoning toggle (thinking=disabled) is a native Chat V2 param and
+            # may not pass through /compatibility/v1 — treat as thinking-on.
+            "command-a-plus-05-2026": "default - newest command a+, 128K context",
+            "command-a-reasoning-08-2025": "reasoning - 256K context, thinking is a toggle",
+            "command-a-03-2025": "advanced - 256K context, prior flagship",
+            "command-r7b-12-2024": "economy - cheapest live chat model, 128K",
+        },
+        model_aliases={
+            "command": "command-a-03-2025",
+            "command-light": "command-r-08-2024",
+            "command-r": "command-r-08-2024",
+            "command-r-plus": "command-r-plus-08-2024",
+        },
+        validation_model="command-r7b-12-2024",
+        web_search_adapter=COHERE_WEB_SEARCH_ADAPTER,
+        setup_warning=(
+            "Cohere shut down the v1 hosted web-search connector on 2025-09-15; "
+            "hosted web search on this preset needs migration to an external "
+            "search adapter."
+        ),
+        notes=(
+            "Chat uses Cohere's OpenAI compatibility API (api.cohere.ai/compatibility/v1 — "
+            "documented and correct; do not migrate to v2/chat). The v1 hosted web-search "
+            "connector this preset's adapter targeted was shut down 2025-09-15."
+        ),
+    ),
+    ProfilePreset(
+        key="openrouter",
+        provider_key="openrouter",
+        label="OpenRouter (gateway)",
+        protocol="openai_compat",
+        base_url="https://openrouter.ai/api/v1",
+        api_key_env="OPENROUTER_API_KEY",
+        suggested_models=(
+            "anthropic/claude-sonnet-5",
+            "anthropic/claude-opus-4.8",
+            "openai/gpt-5.6-terra",
+            "openai/gpt-5.6-luna",
+            "z-ai/glm-5.2",
+            "deepseek/deepseek-v4-pro-0813",
+            "deepseek/deepseek-v4-flash-0731",
+            "deepseek/deepseek-v4-flash-vision-exp",
+            "qwen/qwen3.8-max",
+        ),
+        suggested_model_descriptions={
+            # Vendor prefixes are exact: z-ai/ (not zai/), x-ai/, moonshotai/.
+            # Avoid '-latest' floating aliases and rate-limited :free variants
+            # for agent loops.
+            "anthropic/claude-sonnet-5": "default - coding and agents, 1M context",
+            "anthropic/claude-opus-4.8": "advanced - long-horizon autonomous work",
+            "openai/gpt-5.6-terra": "coding - balanced gpt-5.6 tier, 1.05M context",
+            "openai/gpt-5.6-luna": "fast - cost-efficient gpt-5.6 tier",
+            "z-ai/glm-5.2": "economy - cheap 1M-context tool caller",
+            "deepseek/deepseek-v4-pro-0813": "agentic - current reasoning MoE, 1M context",
+            "deepseek/deepseek-v4-flash-0731": "fast - current low-cost 1M release",
+            "deepseek/deepseek-v4-flash-vision-exp": (
+                "vision preview - image understanding and tools, 1M context"
+            ),
+            "qwen/qwen3.8-max": "multimodal - flagship Qwen agent model, 1M context",
+        },
+        validation_model="deepseek/deepseek-v4-flash-0731",
+        web_search_adapter=OPENROUTER_WEB_ADAPTER,
+        cache_capability=_OPENROUTER_STICKY_SESSION_CACHE_CAPABILITY,
+        setup_warning=(
+            "OpenRouter routes through upstream providers; availability, pricing, privacy, "
+            "and parameter support can vary by route."
+        ),
+        notes="Single API to many providers' models.",
+    ),
+    ProfilePreset(
+        key="perplexity",
+        provider_key="perplexity",
+        label="Perplexity Sonar",
+        protocol="openai_compat",
+        base_url="https://api.perplexity.ai",
+        api_key_env="PERPLEXITY_API_KEY",
+        suggested_models=("sonar-pro", "sonar"),
+        web_search_adapter=PERPLEXITY_SONAR_ADAPTER,
+        web_search_model="sonar",
+        setup_warning=(
+            "Search-only: sonar models reject tool definitions (HTTP 400), so this "
+            "preset cannot run agentic tool loops. Perplexity's coding models live "
+            "on the Agent API (/v1/agent), which needs a Responses-style client "
+            "Alysis Code does not ship yet."
+        ),
+        notes="Sonar models include web-grounded answers and citations.",
+    ),
+    ProfilePreset(
+        key="together",
+        provider_key="together",
+        label="Together AI",
+        protocol="openai_compat",
+        base_url="https://api.together.ai/v1",
+        api_key_env="TOGETHER_API_KEY",
+        suggested_models=(
+            "zai-org/GLM-5.2",
+            "moonshotai/Kimi-K2.7-Code",
+            "deepseek-ai/DeepSeek-V4-Pro-0813",
+            "deepseek-ai/DeepSeek-V4-Flash-0731",
+            "MiniMaxAI/MiniMax-M3",
+            "openai/gpt-oss-120b",
+            "openai/gpt-oss-20b",
+        ),
+        suggested_model_descriptions={
+            # Ids are case-sensitive and vendor-prefixed. Kimi-K2.7-Code and
+            # MiniMax-M3 reason unconditionally — never emit a reasoning-off or
+            # effort param for them.
+            "zai-org/GLM-5.2": "default - general coding, 256K context",
+            "moonshotai/Kimi-K2.7-Code": "coding - code specialist, 256K context",
+            "deepseek-ai/DeepSeek-V4-Pro-0813": "reasoning - current flagship, 1M context",
+            "deepseek-ai/DeepSeek-V4-Flash-0731": "fast - current 1M-context release",
+            "MiniMaxAI/MiniMax-M3": "economy - cheapest 512K-context option",
+            "openai/gpt-oss-120b": "open - larger tool-capable GPT-OSS model, 128K context",
+            "openai/gpt-oss-20b": "fallback - cheapest tool-capable id",
+        },
+        model_aliases={
+            # Fallback policy, NOT vendor renames: Together retires serverless
+            # models with a blank successor column. Two are cross-vendor
+            # substitutions — surface the swap to the user at resolution time.
+            "zai-org/GLM-5.1": "zai-org/GLM-5.2",
+            "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": "moonshotai/Kimi-K2.7-Code",
+            "Qwen/Qwen3-Coder-Next-FP8": "moonshotai/Kimi-K2.7-Code",
+        },
+        validation_model="openai/gpt-oss-20b",
+        setup_warning=(
+            "Together retires serverless models on a published schedule with no "
+            "successor mapping — expect id churn; verify access with Together's "
+            "Models API."
+        ),
+    ),
+    ProfilePreset(
+        key="fireworks",
+        provider_key="fireworks",
+        label="Fireworks AI",
+        protocol="openai_compat",
+        base_url="https://api.fireworks.ai/inference/v1",
+        api_key_env="FIREWORKS_API_KEY",
+        suggested_models=(
+            "accounts/fireworks/models/glm-5p2",
+            "accounts/fireworks/models/kimi-k2p7-code",
+            "accounts/fireworks/models/deepseek-v4-pro-0813",
+            "accounts/fireworks/models/deepseek-v4-flash-0731",
+            "accounts/fireworks/models/minimax-m3",
+            "accounts/fireworks/models/qwen3p7-plus",
+        ),
+        suggested_model_descriptions={
+            # 'p' is the decimal convention (5p2 = 5.2). Catalog membership does
+            # NOT imply serverless availability on Fireworks — every id here is
+            # confirmed serverless-capable.
+            "accounts/fireworks/models/glm-5p2": "default - general agentic coding, 1M context",
+            "accounts/fireworks/models/kimi-k2p7-code": "coding - 262K context, tool calling",
+            "accounts/fireworks/models/deepseek-v4-pro-0813": (
+                "reasoning - current 1M-context production release"
+            ),
+            "accounts/fireworks/models/deepseek-v4-flash-0731": (
+                "fast - current lowest-cost 1M-context release"
+            ),
+            "accounts/fireworks/models/minimax-m3": "economy - 512K context, effort control",
+            "accounts/fireworks/models/qwen3p7-plus": "fallback - 262K context, standard tier only",
+        },
+        model_aliases={
+            # qwen2p5-coder is not serverless-capable at all (on-demand GPU
+            # only); the other two are superseded snapshots.
+            "accounts/fireworks/models/qwen2p5-coder-32b-instruct": (
+                "accounts/fireworks/models/kimi-k2p7-code"
+            ),
+            "accounts/fireworks/models/kimi-k2p6": "accounts/fireworks/models/kimi-k2p7-code",
+            "accounts/fireworks/models/glm-5p1": "accounts/fireworks/models/glm-5p2",
+        },
+        validation_model="accounts/fireworks/models/deepseek-v4-flash-0731",
+    ),
+    # Account-gated hosted preset — registered after the hosted third-party
+    # vendors so listings that read PROFILE_PRESETS order do not headline it.
+    ProfilePreset(
+        key="alysis",
+        provider_key="alysis",
+        label="Alysis Code Pro",
+        protocol="openai_compat",
+        # The Alysis Code hosted proxy (`llm` Supabase Edge Function). It
+        # authenticates the user's slk_ key, meters the free daily allowance /
+        # Pro credits server-side, and forwards to DeepSeek. The login flow
+        # overrides this from alysis_cloud at runtime (env-configurable), so
+        # this literal is just the default.
+        base_url="https://vzigujbcjjmpntxhmyvr.supabase.co/functions/v1/llm/v1",
+        api_key_env=None,
+        # The models the subscription offers. Live availability is discovered
+        # from the gateway's /v1/models at runtime; this static list is the
+        # offline fallback and the menu shown before a model is chosen.
+        suggested_models=("deepseek-v4-flash", "deepseek-v4-pro"),
+        suggested_model_descriptions={
+            "deepseek-v4-flash": "default - fast high-volume coding (1M context, free daily allowance)",
+            "deepseek-v4-pro": "flagship - deeper reasoning (1M context, requires Alysis Code Pro)",
+        },
+        validation_model="deepseek-v4-flash",
+        # Migrate ids from the retired Xiaomi MiMo trial to the Pro default so
+        # old sessions keep working after upgrade.
+        model_aliases={
+            "mimo": "deepseek-v4-flash",
+            "mimo-v2.5-pro": "deepseek-v4-flash",
+            "mimo-v2-flash": "deepseek-v4-flash",
+            "mimo-v2.5": "deepseek-v4-flash",
+        },
+        setup_warning=("Requires an Alysis Code Pro subscription — run `alysis login` to connect."),
+        notes="Hosted models via your Alysis Code Pro subscription. Authenticate with `alysis login`.",
+    ),
+    ProfilePreset(
+        key="ollama",
+        provider_key="ollama",
+        label="Ollama (local)",
+        protocol="openai_compat",
+        base_url="http://localhost:11434/v1",
+        api_key_env=None,
+        suggested_models=("llama3.3",),
+        notes="Local Ollama server. No API key required.",
+    ),
+    ProfilePreset(
+        key="lm-studio",
+        provider_key="lm-studio",
+        label="LM Studio (local)",
+        protocol="openai_compat",
+        base_url="http://localhost:1234/v1",
+        api_key_env=None,
+        suggested_models=("local-model",),
+        notes="Local LM Studio server. No API key required.",
+    ),
+    ProfilePreset(
+        key="vllm",
+        provider_key="vllm",
+        label="vLLM (self-hosted)",
+        protocol="openai_compat",
+        base_url="http://localhost:8000/v1",
+        api_key_env=None,
+        suggested_models=("local-model",),
+    ),
+    ProfilePreset(
+        key="custom",
+        label="Custom (specify URL manually)",
+        protocol="openai_compat",
+        base_url="",
+        api_key_env=None,
+        suggested_models=(),
+        notes="Use for unlisted endpoints. Type the URL during setup.",
+    ),
+)
+
+PRESET_BY_KEY: dict[str, ProfilePreset] = {preset.key: preset for preset in PROFILE_PRESETS}
+
+
+def get_preset(key: str) -> ProfilePreset | None:
+    return PRESET_BY_KEY.get(str(key or "").strip().lower())
+
+
+def model_options_for_preset(preset: ProfilePreset) -> tuple[tuple[str, str, str], ...]:
+    """Return picker rows for the models this preset intentionally supports."""
+    rows: list[tuple[str, str, str]] = []
+    seen: set[str] = set()
+    for model in preset.suggested_models:
+        model_id = str(model or "").strip()
+        if not model_id or model_id in seen:
+            continue
+        seen.add(model_id)
+        description = str(preset.suggested_model_descriptions.get(model_id) or "").strip()
+        rows.append((model_id, model_id, description or "suggested by provider preset"))
+    return tuple(rows)
+
+
+def canonical_model_alias_for_preset(preset: ProfilePreset, model: str) -> str:
+    """Map explicit stale provider aliases to the preset's current model ID."""
+    raw = str(model or "").strip()
+    if not raw:
+        return raw
+    for alias, canonical in preset.model_aliases.items():
+        if str(alias or "").strip().casefold() == raw.casefold():
+            normalized = str(canonical or "").strip()
+            return normalized or raw
+    return raw
+
+
+def find_preset_for_profile(profile: ProfileSpec) -> ProfilePreset | None:
+    """Best-effort mapping from a persisted profile back to a known provider preset."""
+    name_match = get_preset(profile.name)
+    if name_match is not None and _profile_matches_preset(profile, name_match):
+        return name_match
+
+    for preset in PROFILE_PRESETS:
+        if preset.key == "custom":
+            continue
+        if _profile_matches_preset(profile, preset):
+            return preset
+
+    if profile.protocol != OPENAI_COMPAT_PROTOCOL:
+        return None
+    return find_preset_for_base_url(profile.base_url)
+
+
+def find_preset_for_base_url(base_url: str) -> ProfilePreset | None:
+    normalized = _normalized_base_url(base_url)
+    if not normalized:
+        return None
+    matches: list[ProfilePreset] = []
+    for preset in PROFILE_PRESETS:
+        if preset.key == "custom":
+            continue
+        if _normalized_base_url(preset.base_url) == normalized:
+            matches.append(preset)
+    if not matches:
+        return None
+    compatibility = next(
+        (preset for preset in matches if preset.protocol == OPENAI_COMPAT_PROTOCOL),
+        None,
+    )
+    return compatibility or matches[0]
+
+
+def _profile_matches_preset(profile: ProfileSpec, preset: ProfilePreset) -> bool:
+    if str(profile.protocol or OPENAI_COMPAT_PROTOCOL).strip() != preset.protocol:
+        return False
+    profile_url = _normalized_base_url(profile.base_url)
+    preset_url = _normalized_base_url(preset.base_url)
+    if not profile_url or not preset_url or profile_url != preset_url:
+        return False
+    return True
+
+
+def profile_provider_family(profile: ProfileSpec) -> str | None:
+    """Resolve a profile to a first-party family for protocol conversion and diagnostics."""
+    protocol = str(profile.protocol or OPENAI_COMPAT_PROTOCOL).strip()
+    if protocol == OPENAI_RESPONSES_PROTOCOL:
+        return "openai"
+    if protocol == ANTHROPIC_MESSAGES_PROTOCOL:
+        return "anthropic"
+    if protocol == GEMINI_GENERATE_CONTENT_PROTOCOL:
+        return "gemini"
+    if protocol == GEMINI_INTERACTIONS_PROTOCOL:
+        return "gemini"
+
+    preset = get_preset(profile.name)
+    if preset is not None:
+        if preset.key in {"openai", "openai-responses"}:
+            return "openai"
+        if preset.key in {"anthropic", "anthropic-compat", "anthropic-native"}:
+            return "anthropic"
+        if preset.key in {"gemini", "gemini-compat", "gemini-native"}:
+            return "gemini"
+
+    normalized_name = str(profile.name or "").strip().lower()
+    if "openai" in normalized_name:
+        return "openai"
+    if "anthropic" in normalized_name or "claude" in normalized_name:
+        return "anthropic"
+    if "gemini" in normalized_name or "google" in normalized_name:
+        return "gemini"
+
+    parsed = _split_base_url(profile.base_url)
+    if parsed[0] == "api.openai.com":
+        return "openai"
+    if parsed[0] == "api.anthropic.com":
+        return "anthropic"
+    if parsed[0] == "generativelanguage.googleapis.com":
+        return "gemini"
+    return None
+
+
+def target_preset_for_profile_conversion(
+    profile: ProfileSpec,
+    *,
+    target: str,
+) -> ProfilePreset | None:
+    normalized_target = normalize_conversion_target(target)
+    family = profile_provider_family(profile)
+    if family is None:
+        return None
+    preset_key = _CONVERSION_PRESET_BY_FAMILY.get(family, {}).get(normalized_target)
+    if preset_key is None:
+        return None
+    return get_preset(preset_key)
+
+
+def normalize_conversion_target(value: str) -> str:
+    target = str(value or "").strip().lower().replace("_", "-")
+    if target in {"native", "first-party", "firstparty"}:
+        return "native"
+    if target in {"compat", "compatibility", "openai-compatible", "gateway"}:
+        return "compatibility"
+    raise ValueError("conversion target must be 'native' or 'compatibility'")
+
+
+def convert_profile_to_preset(profile: ProfileSpec, preset: ProfilePreset) -> ProfileSpec:
+    current_model = str(profile.default_model or "").strip()
+    default_model = canonical_model_alias_for_preset(preset, current_model)
+    target_family = _preset_provider_family(preset)
+    if not default_model or _model_known_incompatible_with_family(default_model, target_family):
+        default_model = preset.suggested_models[0] if preset.suggested_models else ""
+    notes = _converted_profile_notes(profile, preset)
+
+    return ProfileSpec(
+        name=profile.name,
+        protocol=preset.protocol,
+        base_url=preset.base_url,
+        api_key_env=profile.api_key_env or preset.api_key_env,
+        extra_headers=dict(profile.extra_headers),
+        default_model=default_model,
+        reasoning_effort=profile.reasoning_effort,
+        web_search_adapter=preset.web_search_adapter,
+        web_search_model=preset.web_search_model,
+        notes=notes,
+    )
+
+
+def _converted_profile_notes(profile: ProfileSpec, preset: ProfilePreset) -> str:
+    notes = str(profile.notes or "").strip()
+    if not notes:
+        return preset.notes
+    source_preset = find_preset_for_profile(profile)
+    if source_preset is not None and notes == str(source_preset.notes or "").strip():
+        return preset.notes
+
+    lowered = notes.lower()
+    target_is_native = preset.protocol in NATIVE_PROFILE_PROTOCOLS
+    if target_is_native and (
+        "openai-compat" in lowered
+        or "openai compatible" in lowered
+        or "openai-compatible" in lowered
+        or "compatibility mode" in lowered
+    ):
+        return preset.notes
+    if not target_is_native and "native" in lowered:
+        return preset.notes
+    return notes
+
+
+def _split_base_url(value: str | None) -> tuple[str, str]:
+    try:
+        parsed = urlsplit(str(value or "").strip())
+    except ValueError:
+        return "", ""
+    path = parsed.path.rstrip("/").lower()
+    return (parsed.hostname or "").rstrip(".").lower(), path
+
+
+def _preset_provider_family(preset: ProfilePreset) -> str | None:
+    for family, targets in _CONVERSION_PRESET_BY_FAMILY.items():
+        if preset.key in targets.values():
+            return family
+    return None
+
+
+def _model_known_incompatible_with_family(model: str, family: str | None) -> bool:
+    if family is None:
+        return False
+    normalized = model.strip().lower()
+    model_family = _known_model_family(normalized)
+    if model_family is None:
+        return False
+    if model_family != family:
+        return True
+    return _has_known_provider_namespace(normalized)
+
+
+def known_model_family(model: str) -> str | None:
+    """Best-effort model-family classifier for static diagnostics.
+
+    This is intentionally conservative. Unknown custom gateway models return None so doctor
+    diagnostics do not over-warn on valid provider-specific names Alysis Code cannot know offline.
+    """
+    return _known_model_family(str(model or "").strip().lower())
+
+
+def model_known_incompatible_with_family(model: str, family: str | None) -> bool:
+    """Public wrapper used by diagnostics and tests."""
+    return _model_known_incompatible_with_family(model, family)
+
+
+def _known_model_family(model: str) -> str | None:
+    known_prefixes: dict[str, tuple[str, ...]] = {
+        "openai": ("gpt-", "chatgpt-", "o1", "o3", "o4", "o5"),
+        "anthropic": ("claude-",),
+        "gemini": ("gemini-",),
+    }
+    known_namespaces: dict[str, tuple[str, ...]] = {
+        "openai": ("openai",),
+        "anthropic": ("anthropic", "anthropic-ai"),
+        "gemini": ("google", "gemini"),
+    }
+    parts = [part for part in model.split("/") if part]
+    for known_family, namespaces in known_namespaces.items():
+        if parts and parts[0] in namespaces:
+            return known_family
+    model_id = parts[-1] if parts else model
+    for known_family, prefixes in known_prefixes.items():
+        if model_id.startswith(prefixes):
+            return known_family
+    return None
+
+
+def _has_known_provider_namespace(model: str) -> bool:
+    if "/" not in model:
+        return False
+    namespace = model.split("/", 1)[0]
+    return namespace in {"openai", "anthropic", "anthropic-ai", "google", "gemini"}
+
+
+def _normalized_base_url(value: str | None) -> str:
+    return str(value or "").strip().rstrip("/")
+
+
+def make_profile_from_preset(
+    preset: ProfilePreset,
+    *,
+    name: str | None = None,
+) -> ProfileSpec:
+    profile_name = str(name or preset.key).strip().lower()
+    return ProfileSpec(
+        name=profile_name,
+        protocol=preset.protocol,
+        base_url=preset.base_url,
+        api_key_env=preset.api_key_env,
+        extra_headers=dict(preset.extra_headers),
+        default_model=preset.suggested_models[0] if preset.suggested_models else "",
+        web_search_adapter=preset.web_search_adapter,
+        web_search_model=preset.web_search_model,
+        notes=preset.notes,
+    )